Company Also Will Demonstrate Ultra-Low-Power Computer Vision Applications
Syntiant Corp., a leader in edge AI deployment, today announced that the company will present several of its highly accurate, edge-deployable machine learning models at the NVIDIA GTC developer conference at the San Jose McEnery Convention Center, March 18-21.
Syntiant will demonstrate its optimized LLM in GTC booth I-103 that achieved twice the rate of token generation vs. the SOTA GGML LLaMa-7B benchmark while delivering the same accuracy. These core optimizations reduce the computational footprint of leading LLM architectures, harnessing the power of generative AI while still running cloud-free, at the edge of networks.
Other ML models to be demoed at GTC include: compute-efficient vision, operating at more than 10x the speed and less than 1/10th the memory footprint of typical open-source solutions; and low-power people detection sensing, ideal for in–person detection and person counting, while preserving privacy and maintaining long battery life.
“The ML models we’re demoing at GTC were trained using NVIDIA accelerated computing technologies, which have helped enable AI across many industries,” said Kurt Busch, CEO of Syntiant. “At Syntiant, we’re committed to delivering greater efficiencies as the new interface between humans and machines, allowing customers to quickly benefit from cloud-free advanced intelligence, anywhere, and on any device.”
Syntiant’s hardware-agnostic deep learning models solve critical problems directly on compute-constrained embedded devices, delivering heterogeneous and pervasive solutions that are small, fast and accurate. Optimized to reduce latency and memory footprint, Syntiant’s models are deployable into production on day one and at a lower cost to OEMs.
Click here to register for NVIDIA GTC or visit Syntiant’s virtual booth. Contact [email protected] to arrange a Syntiant demo (Booth I-103) during the conference.
The post Syntiant to Demonstrate 100% Speed-Up of Optimized LLM’s at NVIDIA GTC first appeared on AI-TechPark.
#MachineLearning [Source: AI Techpark]